Two rival coffee chains will open across the street from one another while a mile of road in either direction has none. This looks like a mistake and is close to the opposite.
Traffic is the input being competed for
A retail site is not chosen for the area it covers. It is chosen for the number of people who pass it, and passing volume is wildly uneven along any road.
Intersections, transit stops and the entrances of large stores concentrate movement. A handful of corners in a town carry a large share of everyone's daily path.
Once a location is scarce and identifiable, every chain wants the same few addresses, and they end up beside each other.
Being close captures the undecided customer
Customers with a firm preference will travel. Customers without one buy from whatever is in front of them at the moment the impulse arrives.
Moving away from a competitor concedes the whole pool of undecided buyers near that competitor. Moving adjacent splits it.
The result is that each firm's best individual move is to sit next to the other, even though both would earn more if they agreed to separate.
They use the same data, so they reach the same answer
Site selection at national chains is a modelling exercise. Traffic counts, drive-time catchments, household composition, competitor positions and existing store cannibalisation all feed the decision.
Rival chains buy comparable data and weigh it in comparable ways. When several models are pointed at the same town, they converge on the same short list of parcels.
Clustering is partly just what happens when competitors share a method.
Landlords price the pattern in
Developers understand the dynamic and design for it. A pad site at the corner of a large centre is marketed at a premium precisely because it is fought over.
Lease terms often include exclusivity clauses that block a direct competitor within the same centre, which pushes the rival to the parcel immediately outside it.
That is why the second chain frequently sits just across the property line rather than inside the same lot.
What the pattern leaves out
The same logic explains the gaps. Streets with steady residents but little through movement generate no cluster, because the modelled traffic never justifies the rent.
Retail follows movement rather than population, and the two are not the same map.